TL;DR:
- HKUST Business School introduces InvestLM, an open-source large language model (LLM) tailored for financial generative AI applications.
- InvestLM competes with commercial chatbots like ChatGPT, offering human-like investment-related responses.
- Its model parameters and development insights are publicly accessible, fostering industry collaboration and research.
- Open-source LLMs democratize access to advanced AI, allowing for LLM training with moderate computing resources.
- InvestLM, derived from LLaMA-65B, excels in the financial domain due to instruction fine-tuning.
- Financial experts rate InvestLM’s responses on par with top commercial LLMs, benefiting finance professionals.
- InvestLM enhances work efficiency by providing investment insights and summarizing financial news.
- HKUST promotes in-house LLM development for competitive advantage and data control.
- The project exemplifies HKUST’s leadership in embracing generative AI in education.
- InvestLM’s open-source nature empowers the financial sector and beyond with the transformative potential of generative AI.
Main AI News:
The School of Business and Management at The Hong Kong University of Science and Technology (HKUST Business School) is at the forefront of innovation with the introduction of InvestLM, Hong Kong’s groundbreaking open-source large language model (LLM) designed for financial generative AI (GenAI) applications. This remarkable achievement positions InvestLM as a formidable contender, capable of generating investment-related responses that rival well-known commercial chatbots, including OpenAI’s ChatGPT. In a bold move towards fostering industry collaboration and research advancement, InvestLM’s model parameters and valuable developmental insights have been generously made available to empower industry practitioners and researchers in deploying LLM-based technology.
The AI-powered natural-language chatbots built upon LLMs, boasting billions or even tens of billions of parameters, have long been recognized for their proficiency in handling a vast array of real-time text-generation tasks. Historically, the development of such chat services necessitated substantial computing power, a resource largely exclusive to major corporations. However, the landscape underwent a significant transformation with the advent of open-source general-purpose LLMs earlier this year. This breakthrough democratized access to LLM training, enabling individuals and organizations with moderate computing resources to tailor LLMs to their specific needs.
Drawing inspiration from LLaMA-65B, an open-source general-purpose LLM, the HKUST research team embarked on an ambitious journey to create InvestLM, a state-of-the-art LLM meticulously fine-tuned for the financial domain. Through a technique known as instruction fine-tuning, they harnessed a rich and diverse corpus of finance and investment-related texts, resulting in InvestLM. Remarkably, financial experts, including hedge fund managers and research analysts, have acclaimed InvestLM’s responses as on par with those produced by state-of-the-art commercial LLMs like GPT-3.5, GPT-4, and Claude-2.
This resounding endorsement underscores InvestLM’s exceptional aptitude for comprehending financial texts, promising to significantly enhance the efficiency of finance and investment professionals. Its capabilities extend to providing investment insights, extracting critical information from financial news and reports, and summarizing complex financial data—a game-changer in the financial industry. Notably, when compared to its foundation model, LLaMA-65B, InvestLM exhibits superior control in response generation, eliminating any tendencies towards hallucinations.
Prof. TAM Kan-Yan, Dean of HKUST Business School, emphasized the competitive advantage financial firms can gain by developing LLMs in-house. This approach not only enhances their application of generative AI but also affords better control over proprietary information and customer data. He proudly highlighted HKUST’s leadership in embracing generative AI within the tertiary education sector in Hong Kong. The InvestLM project stands as a testament to the institution’s commitment to providing invaluable insights to the financial sector, enabling them to harness the potential of generative AI while democratizing access to a robust financial LLM.
Prof. YANG Yi, Associate Professor of HKUST’s Department of Information Systems, Business Statistics, and Operations Management, and a key member of the research project team, emphasized the rarity of open-source financial domain LLMs that can deliver insightful responses to investment-related queries. InvestLM fills this void, as validated by financial professionals. By sharing their insights into fine-tuning a foundation model for financial text generation, the HKUST team hopes to empower industry practitioners within the financial sector and beyond, unlocking the transformative power of generative AI.
The research team’s groundbreaking discovery underscores the effectiveness of applying a diverse set of high-quality, domain-specific instructions in training an LLM. This approach substantially enhances the model’s capacity to excel in domain-specific tasks when compared to relying solely on a large volume of general-purpose instructions. In instances where computational resources are constrained, the team has identified that instruction tuning yields the most significant performance improvements for smaller LLMs, providing a practical avenue for resource optimization in LLM development.
Conclusion:
InvestLM’s emergence as an open-source financial language model signifies a significant shift in the financial AI landscape. It empowers financial professionals with advanced text-generation capabilities and offers an opportunity for organizations to develop proprietary AI solutions. The democratization of LLM training is set to catalyze innovation and efficiency within the financial sector, making InvestLM a game-changer in Hong Kong’s financial technology market.
